Abhijnan Chakraborty

The first large-scale audit of the default moderation system on Bluesky is conducted, revealing a human-AI collaborative system where labels for sexual and graphic content are applied automatically in seconds, while nuanced and high stakes labels require more human oversight, taking hours or days.

LLMs can be six times more biased when demographic identity is conveyed implicitly through cultural cues, even when explicit biases are suppressed.